YAKOV KESELMAN

425.638.3968

Yakov.Keselman@gmail.com

Summary

Experienced, highly creative applied researcher is looking for a fast-paced, collaborative environment that will utilize his excellent problem solving and communication skills and expertise in applied data analysis (machine learning, optimization, pattern recognition) and in algorithm and software development.

Relevant Experience

January 2008 – present: Software Development Engineer, Item Data Management, Amazon.com, Seattle, WA.

• Designed and implemented in Python/SQL a rule-based evaluator of approaches to data reconciliation (deriving an optimal value for an attribute from multiple attribute values).
• Designed, implemented in Python/SQL, and validated an approach to data reconciliation.

September 2006 – December 2007: Software Development Engineer, SQL Server Analysis Services, Microsoft Corporation, Redmond, WA.

• Designed, implemented in C++, tested, and documented improved MDX query processing strategies for the next release of Microsoft SQL Server Analysis Services.
• Debugged and fixed several customer-reported issues related to MDX query processing.

September 2005 – August 2006: Independent researcher specializing in applications of machine learning and optimization techniques to practical problems in data analysis.

• Designed and implemented in Java a subsystem aimed at integration of multiple sources of geographical data. The subsystem was used in a larger system that extracts useful geographical patterns from large repositories of satellite imagery (NCSA, NASA).
• Addressed the problem of computing an optimal representative of a cluster of graphs by a combination of several graph-based approximation algorithms. Implemented the approach in C++/LEDA. Validated the approach on real data.
• Reformulated the problem of model-based clustering of image primitives as a learning problem defined on their attributes. Applied existing machine learning techniques to the resulting problem. Obtained preliminary clustering rules, with subsequent validation.

July 2001 – July 2005: Assistant Professor, School of Computer Science, Telecommunications and Information Systems, DePaul University, Chicago, IL.

• Reformulated the problem of image matching as a many-to-many matching problem on attributed graphs. To address the latter problem, adapted existing graph optimization techniques and co-developed novel approximation graph algorithms. Co-implemented the algorithms in C++/LEDA. Validated the approach on real data. Results were published in several leading pattern recognition conferences and journals.
• Supervised a Ph.D. student in designing and implementing evolutionary (genetic) algorithms aimed at finding an optimal artificial neural network architecture for edge detection in images. Results were presented at a student conference.
• Through teaching and research, developed thorough understanding of practical data management techniques, including: indexing and retrieval of textual and image data, system performance optimization via distribution and multithreading, data warehousing, data cleansing, data classification and clustering, data-driven decision making.
• Encouraged student research through projects, seminars, and career-related feedback.
• Submitted several curriculum development proposals addressing students’ needs.

September 1994 – May 2001: Research and Teaching Assistant, Department of Computer Science, Rutgers University, Piscataway, NJ.

• Reformulated the problem of image-based automated model acquisition for generic object recognition as an optimization problem on attributed graphs. Developed approximation algorithms for solving the formulation. Implemented the algorithms in C++/LEDA. Results were published in a leading pattern recognition journal.
• Lead a team of 4 graduate and undergraduate students in the design and development of a distributed multithreaded system for robot tracking and control that is robust to noise and changes in the environment. Implemented system’s GUI in Tcl/Tk and C++.
• Designed and implemented in Matlab a system for extraction and characterization of regions of interest in biomedical images. Validated the system on actual data. Presented an overview of the system at a leading conference in biomedical engineering.
• Developed a model-based approach for tracking polyhedral objects that is robust to noise and occlusion. Implemented the approach in Matlab and validated it on real image sequences. Presented results at a departmental research seminar.
• Implemented in C a distributed version of an optimization algorithm for a linear programming problem on the Parallel Virtual Machine architecture. The implementation was used to obtain results for a journal publication.

List of Publications: http://www.yashma.org/yakovkeselman/Publications/

Education

· Ph.D., Computer Science, Rutgers University, Piscataway, NJ. 3.8 GPA. May, 2005.

o Coursework included artificial intelligence, robust statistical estimation, databases, design and analysis of algorithms, linear programming, randomized algorithms, numerical analysis, network flows, image processing, and computer vision.

o Received Rutgers University fellowships in Digital Libraries and in Cognitive Science.

· M.A., Mathematics, University of Georgia, Athens, GA. 3.8 GPA. June, 1994.

· B.S., Mathematics and Computer Science, the Urals State University, Ekaterinburg, Russia. 3.8 GPA. June, 1991.

Visa Status: US citizen